AI-Assisted DAO Governance: When Should the Community Override the AI?

Hi everyone,

as part of our ongoing development at BlockAI, we are currently working on a new version of an AI-driven DAO. The underlying concept is based on a structured decision-making model known as QOC (Question–Option–Criteria).

We have previously outlined the general idea in a Medium article:

https://blockai.medium.com/qoc-a-new-approach-to-make-daos-more-transparent-and-fair-24821e0c4e71

In the meantime, we have also evaluated this approach scientifically. A preprint of the corresponding paper is available on arXiv:

https://arxiv.org/abs/2511.08641

In the paper, we propose a stepwise introduction of the QOC-based DAO model. Together with Sasha, we have agreed to start with step 2: the AI evaluates and decides on proposals, but its decision is not executed automatically. Instead, the final decision remains with the community via a vote. In this setup, the AI’s output should be understood as a recommendation rather than a binding action.

This has two intended benefits:

  1. It acts as a safety mechanism to verify that the AI behaves as expected.

  2. It allows the community to gradually become familiar with the new decision-making process in a transparent and controllable way.

Within this context, we are currently facing a conceptual challenge. Since AI decisions are not executed directly, but are followed by a community vote, four possible outcome combinations emerge.

If the AI approves a proposal, the situation is straightforward:

  • If the community also approves, the proposal is executed.

  • If the community rejects it, the proposal is not executed.

The situation becomes less clear when the AI rejects a proposal. This raises two key questions:

  1. Should a proposal that is rejected by the AI still be put to a community vote?

  2. If the community votes in favor of such a proposal, should it be executed anyway, or should execution require a positive AI recommendation as well?

We would very much appreciate hearing your perspectives and experiences on how to handle these cases.

Best regards,
Christophe & Marc